A HEAD-AND-FACE ANTHROPOMETRIC SURVEY - The National ...

DATA ANALYSIS
Traditional Measurement Data
The first task in preparing traditional (i.e., measured with tape and calipers)
anthropometric data is to make sure there are no errors. The first line of defense is the
in-field data entry and editing system described in the previous section. Despite the
efficiency of this system, however, erroneous values can sometimes creep into the data
base. Therefore, the data were edited again using a combination of regression and
outlier identification techniques. Demographic data were edited after entry by
examining frequency distributions and identifying unusual values. These values were
changed to “missing” if they could not be verified.
The second task in data preparation of the traditional data was the calculation of the
data weights. Recall that our sampling strategy called for equal representation in each
of the sampling cells. This was done to ensure that we had adequately captured the
anthropometric variability in all segments of the population. People in the work force
do not fall into those cells in equal proportion, however. Therefore, our sample needs
to be proportionately weighted to be accurately representative of the U.S. workforce.
As noted above, we do not have access to demographic statistics for the U.S.
respirator-wearing population, so we weighted to the whole U.S. workforce instead.
Previous anthropometric surveys (e.g., Gordon et al., 1989) used another approach to
correcting the sample demographics to match the total population. In those surveys,
individuals in sampling cells where too many people had been measured were
eliminated from the final data set. This approach allowed for an accurate demographic
representation in the final data set, but had the disadvantage that some sampling cells
(older American Indians, in the Army example) were represented by very, very few
people. The weighting approach we use here does not eliminate any of the measured
subjects. Instead, each measured subject is multiplied so that he/she represents many
other persons with the same demographic characteristics as him/herself.
To calculate the weights, we used the 2000 U.S. census (www.census.gov), and broke
it down into the same categories we had used in our sampling plan. Note our
assumption that the workforce is the total U.S. population between the ages of 18 and
66. Clearly some people in this age range are not in the work force, but we have no
reason to believe the workforce is anthropometrically distinct from the population as a
whole.
The weights are calculated as the relative frequency of a given cell in the Census,
divided by the relative frequency of the same cell in the present study. It can be
expressed as:
[N 1,1 /(N 1,1 + N 1,2 + … + N i,,j )] / [n 1,1 /(n 1,1 + n 1,2 + … + n i,,j )]
where N is the count from the age/race cell in the Census,
n is the count from the age/race cell in the present study,
12